Fast and correctly characterizing tsunami source is crucial to accurate early warning of near-field tsunami. Although there is no way to implement forward quantitative calculation directly, a tsunami source can be inverted using real-time tsunami or seismic waveform records which are detected by Dart buoys, GPS buoys, tidal gauges and seismometers, respectively, and also by joint inversion using different sources data. Because different typical tsunami source models may come to different conclusions, it is important for tsunami early warning and tsunami research to learn about the applicability of different tsunami sources and assess the impacts of tsunami source characteristics differences of near-field tsunami. This study analyzed the effects of six different coseismic fault models on near-field numerical forecasting to the tsunami triggered by the March 11, 2011 Tohoku-oki earthquake. A variable nested algorithm was used to increase spatial resolution in the target region. The finest bathymetric grid resolution was 3 arcsec (approx. 90m). The present work focused on assessing the performance of the finite fault model and uniform slip model in near-field tsunami generation, propagation, inundation and their respective characteristic errors by comparing the simulated data with the measured data. From observed data of the Deep-Ocean Assessment and Reporting for Tsunamis (DART) network, Japan GPS buoys, we selected tide gauges along the coastline of Japan and post-even survey. The measured data were compared with forecasts to assess the sensitivity of the six different sources using error analysis. The results show that the characteristics of the energy distribution of near field tsunami is much dependent on tsunami source geometry. In particular, the strike angle and slip are the most sensitive parameters for the energy distribution of near-field tsunami. Comparison indicates that finite fault models are more reasonable than uniform slip models in fitting maximum tsunami run-up height south of 39 degrees N coastal areas, where occurred the most serious tsunami disaster. A total of 32 sea level monitors including Dart buoys, GPS buoys and tide gauges were used to verify the errors of model data. The simulation results of finite fault models fit the observed records better than uniform slip models as a whole. They have the relatively lower mean absolute/relative error. Fujii's source has the lowest absolute/relative error (0.56 m and 26.71%). UCSB tsunami source also has a better accuracy. At the same time USGSCMT source has the highest precision among three uniform slip models. This paper also suggests that using finite fault models can attain obviously higher precision at tide gauges than uniform slip models with respect to DART buoys and GPS buoys stations, and the errors of tsunami sources have significant orientations. The comparison of tsunami wave spectra was carried out with Fujii's source and UCSB source simulated data. Modeling results from Fujii's source show the better agreement with the spectral energy at wave periods between 12 and 60min than UCSB source. Comparison of tsunami sources inferred from different indirect methods shows the crucial importance of tsunami waveforms for initial seafloor deformation inversion. The joint inversion of tsunami waveform data especially using the deep-ocean tsunami signal can determine the tsunami source quickly and reduce the errors caused by the uncertainty of earthquake rupture processes, which can aid understanding of tsunami generation from earthquakes and nonseismic processes.
